Search

Many challenging problems in linguistic analysis concern structures that have a hybrid character---they show evidence of belonging to two, independently motivated types. Proposals often assign them to one or the other class, requiring complication of the theory to handle their exceptionality. We suggest that there is no satisfactory answer to such conundrums under standard, type-based representational theories, for those theories are founded on discrete topologies. As an alternative, we propose “Metric Grammars”--grammatical systems founded on connected topologies. A metric grammar, a recurrent map that has a neural network at its core, changes its grammatical system slightly with each instance of language experience. Focusing on a grammaticalization episode from the history of English---the development of “sort of" and “kind of" from Noun-Preposition structures into adverbs---we provide evidence that metric grammars exhibit statistical anticipation of categorical change, a phenomenon that is difficult to account for with discrete-topology models.

Formal theories of grammar and traditional parsing models,insofar as they presuppose a categorical notion of grammar,face the challenge of accounting for gradient judgments ofacceptability. This challenge is traditionally met by explaininggradient effects in terms of extra-grammatical factors, positinga purely categorical core for the language system. We presenta new way of accounting for gradience in a self-organizedsentence processing (SOSP) model, which generates structureswith a continuous range of grammaticality values. We focuson islands, a family of syntactic domains out of whichmovement is generally prohibited. Islands are interestingbecause, although most linguistic theories treat them asfully ungrammatical and uninterpretable, experimental studieshave revealed gradient patterns of acceptability and evidencefor their interpretability. We report simulations in whichSOSP largely respects island constraints, but in certain cases,consistent with empirical data, coerces elements that blockdependencies into elements that allow them.

We present a self-organizing approach to sentence processing that sheds new light on notional plurality effects in agreement attraction, using pseudopartitive subject noun phrases (e.g., "a bottle of pills"). We first show that notional plurality ratings (numerosity judgments for subject noun phrases) predict verb agreement choices in pseudopartitives, in line with the "Marking" component of the Marking and Morphing theory of agreement processing. However, no account to date has derived notional plurality values from independently needed principles of language processing. We argue on the basis of new experimental evidence and a dynamical systems model that the theoretical black box of notional plurality can be unpacked into objectively measurable semantic features. With these semantic features driving structure formation (and hence agreement choice), our model reproduces the human verb production patterns as a byproduct of normal processing. Finally, we discuss how the self-organizing approach might be extended to other agreement attraction phenomena.

Gough and Tunmer's (1986) simple view of reading (SVR) proposed that reading comprehension (RC) is a function of language comprehension (LC) and word recognition/decoding. Braze et al. (2007) presented data suggesting an extension of the SVR in which knowledge of vocabulary (V) affected RC over and above the effects of LC. Tunmer and Chapman (2012) found a similar independent contribution of V to RC when the data were analyzed by hierarchical regression. However, additional analysis by factor analysis and structural equation modeling indicated that the effect of V on RC was, in fact, completely captured by LC itself and there was no need to posit a separate direct effect of V on RC. In the present study, we present new data from young adults with sub-optimal reading skill (N = 286). Latent variable and regression analyses support Gough and Tunmer's original proposal and the conclusions of Tunmer and Chapman that V can be considered a component of LC and not an independent contributor to RC.

A dynamical systems model of language processing suggests a resolution of the debate about the influences of syntactic and lexical constraints on processing. Syntactic hypotheses are modeled as attractors which compete for the processor's trajectory. When accumulating evidence puts the processor close to an attractor, processing is quick and lexical differences are hard to detect. When the processor lands between several attractors, multiple hypotheses compete and lexical diiferences can tip the balance one way or the other. This approach allows us to be more explicit about the emergent properties of lexicalist models that are hypothesized to account for syntactic effects (MacDonald, Pearlmutter & Seidenbeig, 1994; Trueswell & Tanenhaus, 1994).

Current, rule-based theories of grammar do not provide much insight into how languages can develop new behaviors over time. Yet, textual data indicate that languages usually evolve new grammatical patterns by gradually extending existing ones. I show how a grammar model that is sensitive to prototype structure can model innovation as a process of extrapolation along salient dimensions of the category clusters. A Connectionist network provides a usefully interpretable implementation. Confirming evidence comes from a study of the development of English be going to as a marker of future tense.

Psycholinguistic research spanning a number of decades has produced diverging results with regard to the nature of constraint integration in online sentence processing. For example, evidence that language users anticipatorily fixate likely upcoming referents in advance of evidence in the speech signal supports rapid context integration. By contrast, evidence that language users activate representations that conflict with contextual constraints, or only indirectly satisfy them, supports nonintegration or late integration. Here we report on a self-organizing neural network framework that addresses 1 aspect of constraint integration: the integration of incoming lexical information (i.e., an incoming word) with sentence context information (i.e., from preceding words in an unfolding utterance). In 2 simulations, we show that the framework predicts both classic results concerned with lexical ambiguity resolution (Swinney, 1979; Tanenhaus, Leiman, & Seidenberg, 1979), which suggest late context integration, and results demonstrating anticipatory eye movements (e.g., Altmann & Kamide, 1999), which support rapid context integration. We also report 2 experiments using the visual world paradigm that confirm a new prediction of the framework. Listeners heard sentences like "The boy will eat the white …" while viewing visual displays with objects like a "white cake" (i.e., a predictable direct object of "eat"), "white car" (i.e., an object not predicted by "eat," but consistent with "white"), and distractors. In line with our simulation predictions, we found that while listeners fixated "white cake" most, they also fixated "white car" more than unrelated distractors in this highly constraining sentence (and visual) context.

Human participants and recurrent ("connectionist") neural networks were both trained on a categorization system abstractly similar to natural language systems involving irregular ("strong") classes and a default class. Both the humans and the networks exhibited staged learning and a generalization pattern reminiscent of the Elsewhere Condition (Kiparsky, 1973). Previous connectionist accounts of related phenomena have often been vague about the nature of the networks' encoding systems. We analyzed our network using dynamical systems theory, revealing topological and geometric properties that can be directly compared with the mechanisms of non-connectionist, rule-based accounts. The results reveal that the networks "contain" structures related to mechanisms posited by rule-based models, partly vindicating the insights of these models. On the other hand, they support the one mechanism (OM), as opposed to the more than one mechanism (MOM), view of symbolic abstraction by showing how the appearance of MOM behavior can arise emergently from one underlying set of principles. The key new contribution of this study is to show that dynamical systems theory can allow us to explicitly characterize the relationship between the two perspectives in implemented models.

This study is part of a broader project aimed at developing cognitive and neurocognitive profiles of adolescent and young adult readers whose educational and occupational prospects are constrained by their limited literacy skills. We explore the relationships among reading-related abilities in participants ages 16 to 24 years spanning a wide range of reading ability. Two specific questions are addressed: (a) Does the simple view of reading capture all nonrandom variation in reading comprehension? (b) Does orally assessed vocabulary knowledge account for variance in reading comprehension, as predicted by the lexical quality hypothesis? A comprehensive battery of cognitive and educational tests was employed to assess phonological awareness, decoding, verbal working memory, listening comprehension, reading comprehension, word knowledge, and experience with print. In this heterogeneous sample, decoding ability clearly played an important role in reading comprehension. The simple view of reading gave a reasonable fit to the data, although it did not capture all of the reliable variance in reading comprehension as predicted. Orally assessed vocabulary knowledge captured unique variance in reading comprehension even after listening comprehension and decoding skill were accounted for. We explore how a specific connectionist model of lexical representation and lexical access can account for these findings.

A central question for psycholinguistics concerns the role of grammatical constraints in online sentence processing. Many current theories maintain that the language processing mechanism constructs a parse or parses that are grammatically consistent with the whole of the perceived input each time it processes a word. Several bottom-up, dynamical models make a contrasting prediction: partial parses which are syntactically compatible with only a proper subpart of the input are sometimes constructed, at least temporarily. Three self-paced reading experiments probed for interference from such locally coherent structures. The first tested for a distracting effect of irrelevant Subject-Predicate interpretations of Noun Phrase-Verb Phrase sequences (e.g., "The coach smiled at the player tossed a frisbee") on reading times. The second addressed the question of whether the interference effects can be treated as lexical interference, instead of involving the formation of locally coherent syntactic structures. The third replicated the reading time effects of the first two experiments with grammaticality judgments. We evaluate the dynamical account, comparing it to other approaches that also predict effects of local coherence, and arguing against accounts which rule out the formation of merely locally coherent structures.

Dynamical, self-organizing models of sentence processing predict "digging-in" effects: The more committed the parser becomes to a wrong syntactic choice, the harder it is to reanalyze. Experiment 1 replicates previous grammaticality judgment studies (F. Ferreira & J. M. Henderson, 1991b, 1993), revealing a deleterious effect of lengthening the ambiguous region of a garden-path sentence. The authors interpret this result as a digging-in effect. Experiment 2 finds a corresponding effect on reading times. Experiment 3 finds that making 2 wrong attachments is worse than making 1. Non-self-organizing models require multiple stipulations to predict both kinds of effects. The authors show that, under an appropriately formulated self-organizing account, both results stem from self-reinforcement of node and link activations, a feature that is needed independently. An implemented model is given.

Proposes a dynamical systems approach to parsing in which syntactic hypotheses are associated with attractors in a metric space. The experiments discussed documented various contingent frequency effects that cut across traditional linguistic grains, each of which was predicted by the dynamical systems model. (47 references) (Author/CK)

The relationship between form and meaning is central to the theory of language. Traditionally, syntax and semantics are viewed as two different levels of representation. Based on insights from the intersection of dynamical systems theory and the theory of computation, and guided by linguistic data, I argue that there is only one space, a syntacticsemantic one. I model it here as a stable, countably infinite attractor of an iterated map dynamical system. One advantage of this approach is that it supports a unified treatment of grammatical and ungrammatical processing. [ABSTRACT FROM AUTHOR]

Training recurrent neural networks on infinite state languages has been successful with languages in which the minimal number of machine states grows linearly with sentence length, but has faired poorly with exponential state-growth languages. A new architecture learns several exponential state-growth languages nearly perfectly by hill climbing.

This paper analyses a three-layer connectionist network that solves a translation-invariance problem, offering a novel explanation for transposed letter effects in word reading. Analysis of the hidden unit encodings provides insight into two central issues in cognitive science: (1) What is the novelty of claims of “modality-specific” encodings? and (2) How can a learning system establish a complex internal structure needed to solve a problem? Although these topics (embodied cognition and learnability) are often treated separately, we find a close relationship between them: modality-specific features help the network discover an abstract encoding by causing it to break the initial symmetries of the hidden units in an effective way. While this neural model is extremely simple compared to the human brain, our results suggest that neural networks need not be black boxes and that carefully examining their encoding behaviours may reveal how they differ from classical ideas about the mind-world relationship. [ABSTRACT FROM AUTHOR]